Data driven design of tonal noise feedback cancellers

Abstract

This paper emphasizes the design methodology for active tonal noise feedback cancellers starting from data collected on the system. To design such control systems, an accurate dynamic model of the system is necessary. Physical modeling can provide qualitative results but fails to yield enough accurate models for control design. The main point in the methodology is concerned by the identification of primary path (noise propagation) and secondary path (compensation) models from data. The procedure is investigated in details starting with transfer functions’ order estimations, continuing with parameters estimation and model’s validation. The second aspect is the design of a noise canceller using the Internal Model Principle and the sensitivity function shaping in order to reduce the ”water-bed” effect. The estimated model’s quality for control design is illustrated by the experimental performance of a tonal noise feedback canceller implemented on a test bench.